Running the R script

Before running the script run_analysis.R, please ensure the following:

1. You have extracted the dataset to a folder called "UCI HAR Dataset" (that has files "features.txt" and "activity_labels.txt", and subfolders "test" and "train").
2. The run_analysis.R file resides at the same level as the folder "UCI HAR Dataset", i.e., both the run_analysis.R file and the folder UCI HAR Dataset belong to the same folder (the run_analysis.R file should not be inside the UCI HAR Dataset folder, for example).
3. On your R console, you have set the working directory (e.g., using setwd()) to the one where both the R script and the UCI HAR Dataset folder reside.

Understanding the R script

The script run_analysis.R is expected to be read/followed along with this description. This description is organized by the steps in the project description - comment lines demarcate the corresponding sections in run_analysis.R.

Step 0: Import Data Sets from Files

Before we can start with any of the steps, we need to import the raw data in the training and test files into R objects. The project data was extracted to a folder called "UCI HAR Dataset" that resides in the same folder as the R script. The path names to the training and test files are constructed in a platform-independent way by using the file.path function.

Step 1: Merging the Training and Test Data

This is done in two steps:

• Combine the columns for subject information (subject), activity label (y), and feature set (X) for training and test data into the data frames raw_train_data and raw_test_data respectively.
• Combine the rows from raw_train_data and raw_test_data to create the single merged data set that is stored in the data frame merged_data.

Step 2: Extracting Mean and Standard Deviation Measurements

This is done in the following steps, corresponding to each uncommented line in the R script:

• The file features.txt contains the feature names (for the columns originally in the "X" files) in its second column. To be able to perform string operations (like checking for inclusion of "mean"), we read in the file as a dataframe measurement_names without the strings (in the second column) being converted into factors.
• We use the names "Subject" and "Activity" for the first two columns (corresponding to the "subject" and "y" data respectively) and the remaining column names (corresponding to the "X" data) are obtained from the second column of measurement_names
• The column names obtained from the file features.txt contain a number of duplicated entries which prevents the select function from working properly. However, these duplicate names do not include the columns of interest (related to mean and standard deviation) and can therefore be first safely removed (using the inbuilt R function duplicated).
• We extract the first two columns and the remaining columns related to mean and standard deviation. For mean, the project instructions may be ambiguous and the justification for my choice is as follows. I have chosen to only extract the mean values of measurements that, based on the description in the file "features_info.txt", are given by names of the form mean() . I agonized long over this, but I decided not to extract mean frequencies (given by names of the form meanFreq()) because following the description in "features_info.txt", they do not directly correspond to means of other measurements in the data, but rather a weighted means of components extracted from the signal measurements. We can unambigously exclude the averaged signals used in the angle() variables since these are angle (and not mean) measurements. The columns of interest related to mean are therefore identified as those containing the substring "mean()" and similarly for standard deviation.

Step 3: Descriptive Activity Names

We will take the descriptive activity names to be those given in the file "activity_labels.txt". The two lines in this section of the R script can be understood as follows:

• We read in the file "activity_labels.txt" as a data frame and use its second column to create the vector activity_names. Note that it so happens that in the file "activity_labels.txt" the sequence of values in the first column (activity labels) are 1, 2, ... and therefore the activity name for any label can be found by simply indexing into the vector activity_names. In other words, for any activity label l (with l ranging from 1 to 6) the corresponding activity name can be obtained as activity_names[l].
• Based on the argument just given, the Activity column of the data frame is mapped to its corresponding descriptive name by simply indexing into the vector activity_names.

Step 4: Adding Descriptive Variable Names

Here, there is a judgement call as to what are suitable variable names. Because of what we did in Step 2, the data frame extracted_data has variable names and these are of the form tBodyAcc-mean()-X, for example. We could chose to rename this to the more verbose possibility "Mean of the Body Acceleration Signal along X-axis" as the descriptive variable name. However, I felt that more succinct names work better as variable names since they appear as column headings and the more verbose description should be deferred to the code book which has a linear presentation. Therefore, given that the variable names I created as part of Step 2 seemed appropriately descriptive to me, I did nothing in this step.

Step 5: Tidy Data Set with Average of each Variable

I have chosen to create the tidy data set in the wide form, where all the measurements (more precisely, their average values) appear in a single row for each subject, activity combination. This satisfies the tidy data requirements since each row contains a single observation (for the combination of subject and activity) and each variable represents a different measurement. This section of the R-script is fairly self-explanatory - main noteworthy uses of the dplyr package are as follows:

• Chaining (%>%) to compose functions
• Using the function summarise_each, rather than summarise, to take the mean of all columns (other than those that are grouped-by) without having to explicitly list them all as would have been required if we had used summarise.

Finally, since the values are now averages per group, we update the variable names in the data frame tidy_averages to more accurately reflect their values (as befitting a "tidy" data set). This is done by prepending the string "Avg:" to all column names other than the first two (corresponding to Subject and Activity) which do not need to be changed.

Coda: Writing the tidy data set to a text file

1. Following the course project instructions, we output the tidy data frame to a text file by using write.table with parameter row.name = FALSE. The output file name has been chosen to be "Tidy Averages.txt". I prefer the look of the text file where strings don't appear within double quotes and therefore have used the parameter quote = FALSE.

2. Note that the text file may not be so easy to read directly. You could use the following sequence of commands to read in the file into R and view it as a nicely formatted data frame (assuming you store the output text file as "Tidy Averages.txt"):

   Tidy.Averages <- read.table("Tidy Averages.txt", header = TRUE, check.names = FALSE)
   View(Tidy.Averages)

Note especially the fact that you should use the parameters header = TRUE (to read the first line as column names) and check.names = FALSE (to prevent R from changing the column names).

3. The more verbose and comprehensive descriptions of the variable names are in the codebook. The codebook is the file "Codebook.txt" in this GIT repository.